Since smartphones have disrupted our daily lives, autonomous driving has become another technology that has disrupted the way we travel. With the reform of the automobile industry to the new four modernizations (electricity, intelligence, networking, and sharing), the complexity of the software and algorithms used in smart cars has grown rapidly. The hardware quality requirements for processing software data and algorithms are getting higher and higher, and the core of the hardware is still the autonomous driving chip.

The autonomous driving chip is essentially a high computing power function chip including a processor and a controller. Currently, the commercial autonomous driving chips that have been mass-produced are basically in the advanced driving assistance system stage, which can realize L1-L2 level assisted driving. The chips used in the quasi-autonomous driving domain controller do not need too high AI computing power, and more logic computing power is used to perform corresponding logic operations.

The chip is like the neural transmission system of the brain in autonomous driving, transmitting every data and algorithm feedback. However, for such important hardware, most of China's smart car field currently relies on chips imported from the United States such as Nvidia and AMD. According to the Global Times, Chinese car companies and autonomous driving companies that focus on autonomous driving technology, such as NIO, Xpeng, Li Auto, Weltmeister, SAIC Motor, Pixel-J and Pony.ai, all clearly use AI chips provided by NVIDIA. In China, Nvidia's Drive Orin AI chip has become a core component of assisted driving technology for electric car makers, and some automakers are also using Nvidia's Xavier chips. However, the U.S. Department of Commerce recently issued an export ban to China on chip manufacturers such as Nvidia and AMD. Will this policy break the pattern of China and the United States leading the autonomous driving industry?

The automotive AI chip industry has high requirements for chip and AI solution providers, and needs to have three capabilities:

1) Strong neural network algorithm ability, the core of intelligent car main control chip is the design of neural network unit;

2) The computing unit design of autonomous vehicles needs to consider issues such as computing power, power consumption and volume, and at the same time achieve chip design and algorithm optimization to maximize effective computing power;

3) Ability to provide "software and hardware integration" platform-level solutions, it is necessary to establish a sufficient open ecosystem, using manufacturers to conduct secondary development, and provide customers with solutions such as perception, mapping, and driving strategies.

The U.S. restrictions on the A100 and H100 exported by Nvidia have a huge impact on AI algorithm training resources including intelligent driving algorithm training, especially for areas such as autonomous driving that require high AI computing power. Even if this is a challenge, it is also an opportunity.

This "core limit order" has also prompted many Chinese companies to "make cores" after cutting off the source of imports. A number of leading companies such as Huawei, Horizon, Baidu, and Black Sesame have invested in the chip manufacturing industry. Horizon has become the first smart cockpit and ADAS chip provider in China to achieve official commercial mass production, providing AI chips with the ultimate energy efficiency ratio through "soft and hard combination" capabilities. At the end of May, Jianghuai Automobile Group and Black Sesame Intelligent reached a mass production cooperation. JAC Group's Sihao brand models will be equipped with Black Sesame Intelligent's Huashan No. 2 A1000 series automatic driving chip to provide computing power support for its L2-level automatic driving system; previously, BYD and Ziyoujia announced that they will be equipped with the Horizon Journey 5 chip. At the same time, Baidu is also working with the domestic AI chip company Horizon to build autonomous driving solutions based on domestic chips. In August, Biren Technology released the first general-purpose GPU chip BR100. Officially, the BR100 set a global computing power record, with 16-bit floating-point computing power reaching over 1000T, 8-bit fixed-point computing power reaching over 2000T, and a single-chip peak computing power reaching PFLOPS level.

In addition to the above-mentioned computing power autonomous driving chips and AI algorithms, there is a third factor that is very important and affects the development of autonomous driving - data. Liu Yanyan, solution architect of Alibaba Cloud, said, "Autonomous driving relies on algorithms for environmental perception, vehicle positioning, path planning, and decision-making control. To make autonomous driving more mature, it is necessary to 'feed' massive scene data to repeatedly train the algorithm. "Data-driven is the trend of autonomous driving development, and digesting massive data relies on huge computing power. As the world's leading data solution provider, Magic Data has a large number of data sets related to intelligent driving, and has automatic driving data in various scenarios. At the same time, it has the capabilities of data collection, labeling, storage, management, training, and cleaning. A sample of its data is as follows:

Malay In-Vehicle Scripted Speech Corpus

Filipino English In-Vehicle Scripted Speech Corpus—Smart Mobility

In-Vehicle Noise Corpus

For more information, contact business@magicdatatech.com.